PGMPool.cpp@ 43363

Last change on this file since 43363 was 43163, checked in by vboxsync, 12 years ago
VMM: A wee bit page pool paranoia related to #6349.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 53.9 KB

Line
1	/* $Id: PGMPool.cpp 43163 2012-09-04 14:12:53Z vboxsync $ */
2	/** @file
3	* PGM Shadow Page Pool.
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/** @page pg_pgm_pool PGM Shadow Page Pool
19	*
20	* Motivations:
21	* -# Relationship between shadow page tables and physical guest pages. This
22	* should allow us to skip most of the global flushes now following access
23	* handler changes. The main expense is flushing shadow pages.
24	* -# Limit the pool size if necessary (default is kind of limitless).
25	* -# Allocate shadow pages from RC. We use to only do this in SyncCR3.
26	* -# Required for 64-bit guests.
27	* -# Combining the PD cache and page pool in order to simplify caching.
28	*
29	*
30	* @section sec_pgm_pool_outline Design Outline
31	*
32	* The shadow page pool tracks pages used for shadowing paging structures (i.e.
33	* page tables, page directory, page directory pointer table and page map
34	* level-4). Each page in the pool has an unique identifier. This identifier is
35	* used to link a guest physical page to a shadow PT. The identifier is a
36	* non-zero value and has a relativly low max value - say 14 bits. This makes it
37	* possible to fit it into the upper bits of the of the aHCPhys entries in the
38	* ram range.
39	*
40	* By restricting host physical memory to the first 48 bits (which is the
41	* announced physical memory range of the K8L chip (scheduled for 2008)), we
42	* can safely use the upper 16 bits for shadow page ID and reference counting.
43	*
44	* Update: The 48 bit assumption will be lifted with the new physical memory
45	* management (PGMPAGE), so we won't have any trouble when someone stuffs 2TB
46	* into a box in some years.
47	*
48	* Now, it's possible for a page to be aliased, i.e. mapped by more than one PT
49	* or PD. This is solved by creating a list of physical cross reference extents
50	* when ever this happens. Each node in the list (extent) is can contain 3 page
51	* pool indexes. The list it self is chained using indexes into the paPhysExt
52	* array.
53	*
54	*
55	* @section sec_pgm_pool_life Life Cycle of a Shadow Page
56	*
57	* -# The SyncPT function requests a page from the pool.
58	* The request includes the kind of page it is (PT/PD, PAE/legacy), the
59	* address of the page it's shadowing, and more.
60	* -# The pool responds to the request by allocating a new page.
61	* When the cache is enabled, it will first check if it's in the cache.
62	* Should the pool be exhausted, one of two things can be done:
63	* -# Flush the whole pool and current CR3.
64	* -# Use the cache to find a page which can be flushed (~age).
65	* -# The SyncPT function will sync one or more pages and insert it into the
66	* shadow PD.
67	* -# The SyncPage function may sync more pages on a later \#PFs.
68	* -# The page is freed / flushed in SyncCR3 (perhaps) and some other cases.
69	* When caching is enabled, the page isn't flush but remains in the cache.
70	*
71	*
72	* @section sec_pgm_pool_impl Monitoring
73	*
74	* We always monitor PAGE_SIZE chunks of memory. When we've got multiple shadow
75	* pages for the same PAGE_SIZE of guest memory (PAE and mixed PD/PT) the pages
76	* sharing the monitor get linked using the iMonitoredNext/Prev. The head page
77	* is the pvUser to the access handlers.
78	*
79	*
80	* @section sec_pgm_pool_impl Implementation
81	*
82	* The pool will take pages from the MM page pool. The tracking data
83	* (attributes, bitmaps and so on) are allocated from the hypervisor heap. The
84	* pool content can be accessed both by using the page id and the physical
85	* address (HC). The former is managed by means of an array, the latter by an
86	* offset based AVL tree.
87	*
88	* Flushing of a pool page means that we iterate the content (we know what kind
89	* it is) and updates the link information in the ram range.
90	*
91	* ...
92	*/
93
94
95	/*******************************************************************************
96	* Header Files *
97	*******************************************************************************/
98	#define LOG_GROUP LOG_GROUP_PGM_POOL
99	#include <VBox/vmm/pgm.h>
100	#include <VBox/vmm/mm.h>
101	#include "PGMInternal.h"
102	#include <VBox/vmm/vm.h>
103	#include "PGMInline.h"
104
105	#include <VBox/log.h>
106	#include <VBox/err.h>
107	#include <iprt/asm.h>
108	#include <iprt/string.h>
109	#include <VBox/dbg.h>
110
111
112	/*******************************************************************************
113	* Internal Functions *
114	*******************************************************************************/
115	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser);
116	#ifdef VBOX_WITH_DEBUGGER
117	static DECLCALLBACK(int) pgmR3PoolCmdCheck(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PVM pVM, PCDBGCVAR paArgs, unsigned cArgs);
118	#endif
119
120	#ifdef VBOX_WITH_DEBUGGER
121	/** Command descriptors. */
122	static const DBGCCMD g_aCmds[] =
123	{
124	/* pszCmd, cArgsMin, cArgsMax, paArgDesc, cArgDescs, fFlags, pfnHandler pszSyntax, ....pszDescription */
125	{ "pgmpoolcheck", 0, 0, NULL, 0, 0, pgmR3PoolCmdCheck, "", "Check the pgm pool pages." },
126	};
127	#endif
128
129	/**
130	* Initializes the pool
131	*
132	* @returns VBox status code.
133	* @param pVM Pointer to the VM.
134	*/
135	int pgmR3PoolInit(PVM pVM)
136	{
137	int rc;
138
139	AssertCompile(NIL_PGMPOOL_IDX == 0);
140	/* pPage->cLocked is an unsigned byte. */
141	AssertCompile(VMM_MAX_CPU_COUNT <= 255);
142
143	/*
144	* Query Pool config.
145	*/
146	PCFGMNODE pCfg = CFGMR3GetChild(CFGMR3GetRoot(pVM), "/PGM/Pool");
147
148	/* Default pgm pool size is 1024 pages (4MB). */
149	uint16_t cMaxPages = 1024;
150
151	/* Adjust it up relative to the RAM size, using the nested paging formula. */
152	uint64_t cbRam;
153	rc = CFGMR3QueryU64Def(CFGMR3GetRoot(pVM), "RamSize", &cbRam, 0); AssertRCReturn(rc, rc);
154	uint64_t u64MaxPages = (cbRam >> 9)
155	+ (cbRam >> 18)
156	+ (cbRam >> 27)
157	+ 32 * PAGE_SIZE;
158	u64MaxPages >>= PAGE_SHIFT;
159	if (u64MaxPages > PGMPOOL_IDX_LAST)
160	cMaxPages = PGMPOOL_IDX_LAST;
161	else
162	cMaxPages = (uint16_t)u64MaxPages;
163
164	/** @cfgm{/PGM/Pool/MaxPages, uint16_t, #pages, 16, 0x3fff, F(ram-size)}
165	* The max size of the shadow page pool in pages. The pool will grow dynamically
166	* up to this limit.
167	*/
168	rc = CFGMR3QueryU16Def(pCfg, "MaxPages", &cMaxPages, cMaxPages);
169	AssertLogRelRCReturn(rc, rc);
170	AssertLogRelMsgReturn(cMaxPages <= PGMPOOL_IDX_LAST && cMaxPages >= RT_ALIGN(PGMPOOL_IDX_FIRST, 16),
171	("cMaxPages=%u (%#x)\n", cMaxPages, cMaxPages), VERR_INVALID_PARAMETER);
172	cMaxPages = RT_ALIGN(cMaxPages, 16);
173	if (cMaxPages > PGMPOOL_IDX_LAST)
174	cMaxPages = PGMPOOL_IDX_LAST;
175	LogRel(("PGMPool: cMaxPages=%u (u64MaxPages=%llu)\n", cMaxPages, u64MaxPages));
176
177	/** todo:
178	* We need to be much more careful with our allocation strategy here.
179	* For nested paging we don't need pool user info nor extents at all, but
180	* we can't check for nested paging here (too early during init to get a
181	* confirmation it can be used). The default for large memory configs is a
182	* bit large for shadow paging, so I've restricted the extent maximum to 8k
183	* (8k * 16 = 128k of hyper heap).
184	*
185	* Also when large page support is enabled, we typically don't need so much,
186	* although that depends on the availability of 2 MB chunks on the host.
187	*/
188
189	/** @cfgm{/PGM/Pool/MaxUsers, uint16_t, #users, MaxUsers, 32K, MaxPages*2}
190	* The max number of shadow page user tracking records. Each shadow page has
191	* zero of other shadow pages (or CR3s) that references it, or uses it if you
192	* like. The structures describing these relationships are allocated from a
193	* fixed sized pool. This configuration variable defines the pool size.
194	*/
195	uint16_t cMaxUsers;
196	rc = CFGMR3QueryU16Def(pCfg, "MaxUsers", &cMaxUsers, cMaxPages * 2);
197	AssertLogRelRCReturn(rc, rc);
198	AssertLogRelMsgReturn(cMaxUsers >= cMaxPages && cMaxPages <= _32K,
199	("cMaxUsers=%u (%#x)\n", cMaxUsers, cMaxUsers), VERR_INVALID_PARAMETER);
200
201	/** @cfgm{/PGM/Pool/MaxPhysExts, uint16_t, #extents, 16, MaxPages * 2, MIN(MaxPages*2,8192)}
202	* The max number of extents for tracking aliased guest pages.
203	*/
204	uint16_t cMaxPhysExts;
205	rc = CFGMR3QueryU16Def(pCfg, "MaxPhysExts", &cMaxPhysExts,
206	RT_MIN(cMaxPages * 2, 8192 /* 8Ki max as this eat too much hyper heap */));
207	AssertLogRelRCReturn(rc, rc);
208	AssertLogRelMsgReturn(cMaxPhysExts >= 16 && cMaxPhysExts <= PGMPOOL_IDX_LAST,
209	("cMaxPhysExts=%u (%#x)\n", cMaxPhysExts, cMaxPhysExts), VERR_INVALID_PARAMETER);
210
211	/** @cfgm{/PGM/Pool/ChacheEnabled, bool, true}
212	* Enables or disabling caching of shadow pages. Caching means that we will try
213	* reuse shadow pages instead of recreating them everything SyncCR3, SyncPT or
214	* SyncPage requests one. When reusing a shadow page, we can save time
215	* reconstructing it and it's children.
216	*/
217	bool fCacheEnabled;
218	rc = CFGMR3QueryBoolDef(pCfg, "CacheEnabled", &fCacheEnabled, true);
219	AssertLogRelRCReturn(rc, rc);
220
221	LogRel(("pgmR3PoolInit: cMaxPages=%#RX16 cMaxUsers=%#RX16 cMaxPhysExts=%#RX16 fCacheEnable=%RTbool\n",
222	cMaxPages, cMaxUsers, cMaxPhysExts, fCacheEnabled));
223
224	/*
225	* Allocate the data structures.
226	*/
227	uint32_t cb = RT_OFFSETOF(PGMPOOL, aPages[cMaxPages]);
228	cb += cMaxUsers * sizeof(PGMPOOLUSER);
229	cb += cMaxPhysExts * sizeof(PGMPOOLPHYSEXT);
230	PPGMPOOL pPool;
231	rc = MMR3HyperAllocOnceNoRel(pVM, cb, 0, MM_TAG_PGM_POOL, (void **)&pPool);
232	if (RT_FAILURE(rc))
233	return rc;
234	pVM->pgm.s.pPoolR3 = pPool;
235	pVM->pgm.s.pPoolR0 = MMHyperR3ToR0(pVM, pPool);
236	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pPool);
237
238	/*
239	* Initialize it.
240	*/
241	pPool->pVMR3 = pVM;
242	pPool->pVMR0 = pVM->pVMR0;
243	pPool->pVMRC = pVM->pVMRC;
244	pPool->cMaxPages = cMaxPages;
245	pPool->cCurPages = PGMPOOL_IDX_FIRST;
246	pPool->iUserFreeHead = 0;
247	pPool->cMaxUsers = cMaxUsers;
248	PPGMPOOLUSER paUsers = (PPGMPOOLUSER)&pPool->aPages[pPool->cMaxPages];
249	pPool->paUsersR3 = paUsers;
250	pPool->paUsersR0 = MMHyperR3ToR0(pVM, paUsers);
251	pPool->paUsersRC = MMHyperR3ToRC(pVM, paUsers);
252	for (unsigned i = 0; i < cMaxUsers; i++)
253	{
254	paUsers[i].iNext = i + 1;
255	paUsers[i].iUser = NIL_PGMPOOL_IDX;
256	paUsers[i].iUserTable = 0xfffffffe;
257	}
258	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
259	pPool->iPhysExtFreeHead = 0;
260	pPool->cMaxPhysExts = cMaxPhysExts;
261	PPGMPOOLPHYSEXT paPhysExts = (PPGMPOOLPHYSEXT)&paUsers[cMaxUsers];
262	pPool->paPhysExtsR3 = paPhysExts;
263	pPool->paPhysExtsR0 = MMHyperR3ToR0(pVM, paPhysExts);
264	pPool->paPhysExtsRC = MMHyperR3ToRC(pVM, paPhysExts);
265	for (unsigned i = 0; i < cMaxPhysExts; i++)
266	{
267	paPhysExts[i].iNext = i + 1;
268	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
269	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
270	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
271	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
272	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
273	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
274	}
275	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
276	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
277	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
278	pPool->iAgeHead = NIL_PGMPOOL_IDX;
279	pPool->iAgeTail = NIL_PGMPOOL_IDX;
280	pPool->fCacheEnabled = fCacheEnabled;
281	pPool->pfnAccessHandlerR3 = pgmR3PoolAccessHandler;
282	pPool->pszAccessHandler = "Guest Paging Access Handler";
283	pPool->HCPhysTree = 0;
284
285	/* The NIL entry. */
286	Assert(NIL_PGMPOOL_IDX == 0);
287	pPool->aPages[NIL_PGMPOOL_IDX].enmKind = PGMPOOLKIND_INVALID;
288	pPool->aPages[NIL_PGMPOOL_IDX].idx = NIL_PGMPOOL_IDX;
289
290	/* The Shadow 32-bit PD. (32 bits guest paging) */
291	pPool->aPages[PGMPOOL_IDX_PD].enmKind = PGMPOOLKIND_32BIT_PD;
292	pPool->aPages[PGMPOOL_IDX_PD].idx = PGMPOOL_IDX_PD;
293
294	/* The Shadow PDPT. */
295	pPool->aPages[PGMPOOL_IDX_PDPT].enmKind = PGMPOOLKIND_PAE_PDPT;
296	pPool->aPages[PGMPOOL_IDX_PDPT].idx = PGMPOOL_IDX_PDPT;
297
298	/* The Shadow AMD64 CR3. */
299	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].enmKind = PGMPOOLKIND_64BIT_PML4;
300	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].idx = PGMPOOL_IDX_AMD64_CR3;
301
302	/* The Nested Paging CR3. */
303	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].enmKind = PGMPOOLKIND_ROOT_NESTED;
304	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].idx = PGMPOOL_IDX_NESTED_ROOT;
305
306	/*
307	* Set common stuff.
308	*/
309	for (unsigned iPage = 0; iPage < PGMPOOL_IDX_FIRST; iPage++)
310	{
311	pPool->aPages[iPage].Core.Key = NIL_RTHCPHYS;
312	pPool->aPages[iPage].GCPhys = NIL_RTGCPHYS;
313	pPool->aPages[iPage].iNext = NIL_PGMPOOL_IDX;
314	/* pPool->aPages[iPage].cLocked = INT32_MAX; - test this out... */
315	pPool->aPages[iPage].pvPageR3 = 0;
316	pPool->aPages[iPage].iUserHead = NIL_PGMPOOL_USER_INDEX;
317	pPool->aPages[iPage].iModifiedNext = NIL_PGMPOOL_IDX;
318	pPool->aPages[iPage].iModifiedPrev = NIL_PGMPOOL_IDX;
319	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
320	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
321	pPool->aPages[iPage].iAgeNext = NIL_PGMPOOL_IDX;
322	pPool->aPages[iPage].iAgePrev = NIL_PGMPOOL_IDX;
323
324	Assert(pPool->aPages[iPage].idx == iPage);
325	Assert(pPool->aPages[iPage].GCPhys == NIL_RTGCPHYS);
326	Assert(!pPool->aPages[iPage].fSeenNonGlobal);
327	Assert(!pPool->aPages[iPage].fMonitored);
328	Assert(!pPool->aPages[iPage].fCached);
329	Assert(!pPool->aPages[iPage].fZeroed);
330	Assert(!pPool->aPages[iPage].fReusedFlushPending);
331	}
332
333	#ifdef VBOX_WITH_STATISTICS
334	/*
335	* Register statistics.
336	*/
337	STAM_REG(pVM, &pPool->cCurPages, STAMTYPE_U16, "/PGM/Pool/cCurPages", STAMUNIT_PAGES, "Current pool size.");
338	STAM_REG(pVM, &pPool->cMaxPages, STAMTYPE_U16, "/PGM/Pool/cMaxPages", STAMUNIT_PAGES, "Max pool size.");
339	STAM_REG(pVM, &pPool->cUsedPages, STAMTYPE_U16, "/PGM/Pool/cUsedPages", STAMUNIT_PAGES, "The number of pages currently in use.");
340	STAM_REG(pVM, &pPool->cUsedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/cUsedPagesHigh", STAMUNIT_PAGES, "The high watermark for cUsedPages.");
341	STAM_REG(pVM, &pPool->StatAlloc, STAMTYPE_PROFILE_ADV, "/PGM/Pool/Alloc", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolAlloc.");
342	STAM_REG(pVM, &pPool->StatClearAll, STAMTYPE_PROFILE, "/PGM/Pool/ClearAll", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolClearAll.");
343	STAM_REG(pVM, &pPool->StatR3Reset, STAMTYPE_PROFILE, "/PGM/Pool/R3Reset", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolReset.");
344	STAM_REG(pVM, &pPool->StatFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushPage.");
345	STAM_REG(pVM, &pPool->StatFree, STAMTYPE_PROFILE, "/PGM/Pool/Free", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFree.");
346	STAM_REG(pVM, &pPool->StatForceFlushPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForce", STAMUNIT_OCCURENCES, "Counting explicit flushes by PGMPoolFlushPage().");
347	STAM_REG(pVM, &pPool->StatForceFlushDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForceDirty", STAMUNIT_OCCURENCES, "Counting explicit flushes of dirty pages by PGMPoolFlushPage().");
348	STAM_REG(pVM, &pPool->StatForceFlushReused, STAMTYPE_COUNTER, "/PGM/Pool/FlushReused", STAMUNIT_OCCURENCES, "Counting flushes for reused pages.");
349	STAM_REG(pVM, &pPool->StatZeroPage, STAMTYPE_PROFILE, "/PGM/Pool/ZeroPage", STAMUNIT_TICKS_PER_CALL, "Profiling time spent zeroing pages. Overlaps with Alloc.");
350	STAM_REG(pVM, &pPool->cMaxUsers, STAMTYPE_U16, "/PGM/Pool/Track/cMaxUsers", STAMUNIT_COUNT, "Max user tracking records.");
351	STAM_REG(pVM, &pPool->cPresent, STAMTYPE_U32, "/PGM/Pool/Track/cPresent", STAMUNIT_COUNT, "Number of present page table entries.");
352	STAM_REG(pVM, &pPool->StatTrackDeref, STAMTYPE_PROFILE, "/PGM/Pool/Track/Deref", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackDeref.");
353	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPT, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPT", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPT.");
354	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTs, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTs", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTs.");
355	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTsSlow, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTsSlow", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTsSlow.");
356	STAM_REG(pVM, &pPool->StatTrackFlushEntry, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Flush", STAMUNIT_COUNT, "Nr of flushed entries.");
357	STAM_REG(pVM, &pPool->StatTrackFlushEntryKeep, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Update", STAMUNIT_COUNT, "Nr of updated entries.");
358	STAM_REG(pVM, &pPool->StatTrackFreeUpOneUser, STAMTYPE_COUNTER, "/PGM/Pool/Track/FreeUpOneUser", STAMUNIT_TICKS_PER_CALL, "The number of times we were out of user tracking records.");
359	STAM_REG(pVM, &pPool->StatTrackDerefGCPhys, STAMTYPE_PROFILE, "/PGM/Pool/Track/DrefGCPhys", STAMUNIT_TICKS_PER_CALL, "Profiling deref activity related tracking GC physical pages.");
360	STAM_REG(pVM, &pPool->StatTrackLinearRamSearches, STAMTYPE_COUNTER, "/PGM/Pool/Track/LinearRamSearches", STAMUNIT_OCCURENCES, "The number of times we had to do linear ram searches.");
361	STAM_REG(pVM, &pPool->StamTrackPhysExtAllocFailures,STAMTYPE_COUNTER, "/PGM/Pool/Track/PhysExtAllocFailures", STAMUNIT_OCCURENCES, "The number of failing pgmPoolTrackPhysExtAlloc calls.");
362	STAM_REG(pVM, &pPool->StatMonitorRZ, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access handler.");
363	STAM_REG(pVM, &pPool->StatMonitorRZEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
364	STAM_REG(pVM, &pPool->StatMonitorRZFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling the pgmPoolFlushPage calls made from the RC/R0 access handler.");
365	STAM_REG(pVM, &pPool->StatMonitorRZFlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
366	STAM_REG(pVM, &pPool->StatMonitorRZFlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
367	STAM_REG(pVM, &pPool->StatMonitorRZFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
368	STAM_REG(pVM, &pPool->StatMonitorRZHandled, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/Handled", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access we've handled (except REP STOSD).");
369	STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch1, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch1", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction.");
370	STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch2, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch2", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction during flushing.");
371	STAM_REG(pVM, &pPool->StatMonitorRZRepPrefix, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/RepPrefix", STAMUNIT_OCCURENCES, "The number of times we've seen rep prefixes we can't handle.");
372	STAM_REG(pVM, &pPool->StatMonitorRZRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
373	STAM_REG(pVM, &pPool->StatMonitorRZFaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
374	STAM_REG(pVM, &pPool->StatMonitorRZFaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
375	STAM_REG(pVM, &pPool->StatMonitorRZFaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
376	STAM_REG(pVM, &pPool->StatMonitorRZFaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
377	STAM_REG(pVM, &pPool->StatMonitorR3, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access handler.");
378	STAM_REG(pVM, &pPool->StatMonitorR3EmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
379	STAM_REG(pVM, &pPool->StatMonitorR3FlushPage, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling the pgmPoolFlushPage calls made from the R3 access handler.");
380	STAM_REG(pVM, &pPool->StatMonitorR3FlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
381	STAM_REG(pVM, &pPool->StatMonitorR3FlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
382	STAM_REG(pVM, &pPool->StatMonitorR3Fork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
383	STAM_REG(pVM, &pPool->StatMonitorR3Handled, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/Handled", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access we've handled (except REP STOSD).");
384	STAM_REG(pVM, &pPool->StatMonitorR3RepPrefix, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/RepPrefix", STAMUNIT_OCCURENCES, "The number of times we've seen rep prefixes we can't handle.");
385	STAM_REG(pVM, &pPool->StatMonitorR3RepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
386	STAM_REG(pVM, &pPool->StatMonitorR3FaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
387	STAM_REG(pVM, &pPool->StatMonitorR3FaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
388	STAM_REG(pVM, &pPool->StatMonitorR3FaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
389	STAM_REG(pVM, &pPool->StatMonitorR3FaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
390	STAM_REG(pVM, &pPool->StatMonitorR3Async, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Async", STAMUNIT_OCCURENCES, "Times we're called in an async thread and need to flush.");
391	STAM_REG(pVM, &pPool->cModifiedPages, STAMTYPE_U16, "/PGM/Pool/Monitor/cModifiedPages", STAMUNIT_PAGES, "The current cModifiedPages value.");
392	STAM_REG(pVM, &pPool->cModifiedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/Monitor/cModifiedPagesHigh", STAMUNIT_PAGES, "The high watermark for cModifiedPages.");
393	STAM_REG(pVM, &pPool->StatResetDirtyPages, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Resets", STAMUNIT_OCCURENCES, "Times we've called pgmPoolResetDirtyPages (and there were dirty page).");
394	STAM_REG(pVM, &pPool->StatDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Pages", STAMUNIT_OCCURENCES, "Times we've called pgmPoolAddDirtyPage.");
395	STAM_REG(pVM, &pPool->StatDirtyPageDupFlush, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/FlushDup", STAMUNIT_OCCURENCES, "Times we've had to flush duplicates for dirty page management.");
396	STAM_REG(pVM, &pPool->StatDirtyPageOverFlowFlush, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/FlushOverflow",STAMUNIT_OCCURENCES, "Times we've had to flush because of overflow.");
397	STAM_REG(pVM, &pPool->StatCacheHits, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Hits", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls satisfied by the cache.");
398	STAM_REG(pVM, &pPool->StatCacheMisses, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Misses", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls not statisfied by the cache.");
399	STAM_REG(pVM, &pPool->StatCacheKindMismatches, STAMTYPE_COUNTER, "/PGM/Pool/Cache/KindMismatches", STAMUNIT_OCCURENCES, "The number of shadow page kind mismatches. (Better be low, preferably 0!)");
400	STAM_REG(pVM, &pPool->StatCacheFreeUpOne, STAMTYPE_COUNTER, "/PGM/Pool/Cache/FreeUpOne", STAMUNIT_OCCURENCES, "The number of times the cache was asked to free up a page.");
401	STAM_REG(pVM, &pPool->StatCacheCacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Cacheable", STAMUNIT_OCCURENCES, "The number of cacheable allocations.");
402	STAM_REG(pVM, &pPool->StatCacheUncacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Uncacheable", STAMUNIT_OCCURENCES, "The number of uncacheable allocations.");
403	#endif /* VBOX_WITH_STATISTICS */
404
405	#ifdef VBOX_WITH_DEBUGGER
406	/*
407	* Debugger commands.
408	*/
409	static bool s_fRegisteredCmds = false;
410	if (!s_fRegisteredCmds)
411	{
412	rc = DBGCRegisterCommands(&g_aCmds[0], RT_ELEMENTS(g_aCmds));
413	if (RT_SUCCESS(rc))
414	s_fRegisteredCmds = true;
415	}
416	#endif
417
418	return VINF_SUCCESS;
419	}
420
421
422	/**
423	* Relocate the page pool data.
424	*
425	* @param pVM Pointer to the VM.
426	*/
427	void pgmR3PoolRelocate(PVM pVM)
428	{
429	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3);
430	pVM->pgm.s.pPoolR3->pVMRC = pVM->pVMRC;
431	pVM->pgm.s.pPoolR3->paUsersRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paUsersR3);
432	pVM->pgm.s.pPoolR3->paPhysExtsRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paPhysExtsR3);
433	int rc = PDMR3LdrGetSymbolRC(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerRC);
434	AssertReleaseRC(rc);
435	/* init order hack. */
436	if (!pVM->pgm.s.pPoolR3->pfnAccessHandlerR0)
437	{
438	rc = PDMR3LdrGetSymbolR0(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerR0);
439	AssertReleaseRC(rc);
440	}
441	}
442
443
444	/**
445	* Grows the shadow page pool.
446	*
447	* I.e. adds more pages to it, assuming that hasn't reached cMaxPages yet.
448	*
449	* @returns VBox status code.
450	* @param pVM Pointer to the VM.
451	*/
452	VMMR3DECL(int) PGMR3PoolGrow(PVM pVM)
453	{
454	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
455	AssertReturn(pPool->cCurPages < pPool->cMaxPages, VERR_PGM_POOL_MAXED_OUT_ALREADY);
456
457	/* With 32-bit guests and no EPT, the CR3 limits the root pages to low
458	(below 4 GB) memory. */
459	/** @todo change the pool to handle ROOT page allocations specially when
460	* required. */
461	bool fCanUseHighMemory = HWACCMIsNestedPagingActive(pVM)
462	&& HWACCMGetShwPagingMode(pVM) == PGMMODE_EPT;
463
464	pgmLock(pVM);
465
466	/*
467	* How much to grow it by?
468	*/
469	uint32_t cPages = pPool->cMaxPages - pPool->cCurPages;
470	cPages = RT_MIN(PGMPOOL_CFG_MAX_GROW, cPages);
471	LogFlow(("PGMR3PoolGrow: Growing the pool by %d (%#x) pages. fCanUseHighMemory=%RTbool\n", cPages, cPages, fCanUseHighMemory));
472
473	for (unsigned i = pPool->cCurPages; cPages-- > 0; i++)
474	{
475	PPGMPOOLPAGE pPage = &pPool->aPages[i];
476
477	if (fCanUseHighMemory)
478	pPage->pvPageR3 = MMR3PageAlloc(pVM);
479	else
480	pPage->pvPageR3 = MMR3PageAllocLow(pVM);
481	if (!pPage->pvPageR3)
482	{
483	Log(("We're out of memory!! i=%d fCanUseHighMemory=%RTbool\n", i, fCanUseHighMemory));
484	pgmUnlock(pVM);
485	return i ? VINF_SUCCESS : VERR_NO_PAGE_MEMORY;
486	}
487	pPage->Core.Key = MMPage2Phys(pVM, pPage->pvPageR3);
488	AssertFatal(pPage->Core.Key < _4G \|\| fCanUseHighMemory);
489	pPage->GCPhys = NIL_RTGCPHYS;
490	pPage->enmKind = PGMPOOLKIND_FREE;
491	pPage->idx = pPage - &pPool->aPages[0];
492	LogFlow(("PGMR3PoolGrow: insert page #%#x - %RHp\n", pPage->idx, pPage->Core.Key));
493	pPage->iNext = pPool->iFreeHead;
494	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
495	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
496	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
497	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
498	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
499	pPage->iAgeNext = NIL_PGMPOOL_IDX;
500	pPage->iAgePrev = NIL_PGMPOOL_IDX;
501	/* commit it */
502	bool fRc = RTAvloHCPhysInsert(&pPool->HCPhysTree, &pPage->Core); Assert(fRc); NOREF(fRc);
503	pPool->iFreeHead = i;
504	pPool->cCurPages = i + 1;
505	}
506
507	pgmUnlock(pVM);
508	Assert(pPool->cCurPages <= pPool->cMaxPages);
509	return VINF_SUCCESS;
510	}
511
512
513
514	/**
515	* Worker used by pgmR3PoolAccessHandler when it's invoked by an async thread.
516	*
517	* @param pPool The pool.
518	* @param pPage The page.
519	*/
520	static DECLCALLBACK(void) pgmR3PoolFlushReusedPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
521	{
522	/* for the present this should be safe enough I think... */
523	pgmLock(pPool->pVMR3);
524	if ( pPage->fReusedFlushPending
525	&& pPage->enmKind != PGMPOOLKIND_FREE)
526	pgmPoolFlushPage(pPool, pPage);
527	pgmUnlock(pPool->pVMR3);
528	}
529
530
531	/**
532	* \#PF Handler callback for PT write accesses.
533	*
534	* The handler can not raise any faults, it's mainly for monitoring write access
535	* to certain pages.
536	*
537	* @returns VINF_SUCCESS if the handler has carried out the operation.
538	* @returns VINF_PGM_HANDLER_DO_DEFAULT if the caller should carry out the access operation.
539	* @param pVM Pointer to the VM.
540	* @param GCPhys The physical address the guest is writing to.
541	* @param pvPhys The HC mapping of that address.
542	* @param pvBuf What the guest is reading/writing.
543	* @param cbBuf How much it's reading/writing.
544	* @param enmAccessType The access type.
545	* @param pvUser User argument.
546	*/
547	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf,
548	PGMACCESSTYPE enmAccessType, void *pvUser)
549	{
550	STAM_PROFILE_START(&pVM->pgm.s.pPoolR3->StatMonitorR3, a);
551	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
552	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
553	PVMCPU pVCpu = VMMGetCpu(pVM);
554	LogFlow(("pgmR3PoolAccessHandler: GCPhys=%RGp %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
555	GCPhys, pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
556
557	NOREF(pvBuf); NOREF(enmAccessType);
558
559	/*
560	* We don't have to be very sophisticated about this since there are relativly few calls here.
561	* However, we must try our best to detect any non-cpu accesses (disk / networking).
562	*
563	* Just to make life more interesting, we'll have to deal with the async threads too.
564	* We cannot flush a page if we're in an async thread because of REM notifications.
565	*/
566	pgmLock(pVM);
567	if (PHYS_PAGE_ADDRESS(GCPhys) != PHYS_PAGE_ADDRESS(pPage->GCPhys))
568	{
569	/* Pool page changed while we were waiting for the lock; ignore. */
570	Log(("CPU%d: pgmR3PoolAccessHandler pgm pool page for %RGp changed (to %RGp) while waiting!\n", pVCpu->idCpu, PHYS_PAGE_ADDRESS(GCPhys), PHYS_PAGE_ADDRESS(pPage->GCPhys)));
571	pgmUnlock(pVM);
572	return VINF_PGM_HANDLER_DO_DEFAULT;
573	}
574
575	Assert(pPage->enmKind != PGMPOOLKIND_FREE);
576
577	/* @todo this code doesn't make any sense. remove the if (!pVCpu) block */
578	if (!pVCpu) /** @todo This shouldn't happen any longer, all access handlers will be called on an EMT. All ring-3 handlers, except MMIO, already own the PGM lock. @bugref{3170} */
579	{
580	Log(("pgmR3PoolAccessHandler: async thread, requesting EMT to flush the page: %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
581	pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
582	STAM_COUNTER_INC(&pPool->StatMonitorR3Async);
583	if (!pPage->fReusedFlushPending)
584	{
585	pgmUnlock(pVM);
586	int rc = VMR3ReqCallVoidNoWait(pPool->pVMR3, VMCPUID_ANY, (PFNRT)pgmR3PoolFlushReusedPage, 2, pPool, pPage);
587	AssertRCReturn(rc, rc);
588	pgmLock(pVM);
589	pPage->fReusedFlushPending = true;
590	pPage->cModifications += 0x1000;
591	}
592
593	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
594	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
595	while (cbBuf > 4)
596	{
597	cbBuf -= 4;
598	pvPhys = (uint8_t *)pvPhys + 4;
599	GCPhys += 4;
600	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
601	}
602	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
603	}
604	else if ( ( pPage->cModifications < 96 /* it's cheaper here. */
605	\|\| pgmPoolIsPageLocked(pPage)
606	)
607	&& cbBuf <= 4)
608	{
609	/* Clear the shadow entry. */
610	if (!pPage->cModifications++)
611	pgmPoolMonitorModifiedInsert(pPool, pPage);
612	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
613	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
614	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
615	}
616	else
617	{
618	pgmPoolMonitorChainFlush(pPool, pPage); /* ASSUME that VERR_PGM_POOL_CLEARED can be ignored here and that FFs will deal with it in due time. */
619	STAM_PROFILE_STOP_EX(&pPool->StatMonitorR3, &pPool->StatMonitorR3FlushPage, a);
620	}
621	pgmUnlock(pVM);
622	return VINF_PGM_HANDLER_DO_DEFAULT;
623	}
624
625
626	/**
627	* Rendezvous callback used by pgmR3PoolClearAll that clears all shadow pages
628	* and all modification counters.
629	*
630	* This is only called on one of the EMTs while the other ones are waiting for
631	* it to complete this function.
632	*
633	* @returns VINF_SUCCESS (VBox strict status code).
634	* @param pVM Pointer to the VM.
635	* @param pVCpu The VMCPU for the EMT we're being called on. Unused.
636	* @param fpvFlushRemTlb When not NULL, we'll flush the REM TLB as well.
637	* (This is the pvUser, so it has to be void *.)
638	*
639	*/
640	DECLCALLBACK(VBOXSTRICTRC) pgmR3PoolClearAllRendezvous(PVM pVM, PVMCPU pVCpu, void *fpvFlushRemTbl)
641	{
642	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
643	STAM_PROFILE_START(&pPool->StatClearAll, c);
644	NOREF(pVCpu);
645
646	pgmLock(pVM);
647	Log(("pgmR3PoolClearAllRendezvous: cUsedPages=%d fpvFlushRemTbl=%RTbool\n", pPool->cUsedPages, !!fpvFlushRemTbl));
648
649	/*
650	* Iterate all the pages until we've encountered all that are in use.
651	* This is a simple but not quite optimal solution.
652	*/
653	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
654	unsigned cLeft = pPool->cUsedPages;
655	uint32_t iPage = pPool->cCurPages;
656	while (--iPage >= PGMPOOL_IDX_FIRST)
657	{
658	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
659	if (pPage->GCPhys != NIL_RTGCPHYS)
660	{
661	switch (pPage->enmKind)
662	{
663	/*
664	* We only care about shadow page tables that reference physical memory
665	*/
666	#ifdef PGM_WITH_LARGE_PAGES
667	case PGMPOOLKIND_EPT_PD_FOR_PHYS: /* Large pages reference 2 MB of physical memory, so we must clear them. */
668	if (pPage->cPresent)
669	{
670	PX86PDPAE pShwPD = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
671	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
672	{
673	if ( pShwPD->a[i].n.u1Present
674	&& pShwPD->a[i].b.u1Size)
675	{
676	Assert(!(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING));
677	pShwPD->a[i].u = 0;
678	Assert(pPage->cPresent);
679	pPage->cPresent--;
680	}
681	}
682	if (pPage->cPresent == 0)
683	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
684	}
685	goto default_case;
686
687	case PGMPOOLKIND_PAE_PD_PHYS: /* Large pages reference 2 MB of physical memory, so we must clear them. */
688	if (pPage->cPresent)
689	{
690	PEPTPD pShwPD = (PEPTPD)PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
691	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
692	{
693	Assert((pShwPD->a[i].u & UINT64_C(0xfff0000000000f80)) == 0);
694	if ( pShwPD->a[i].n.u1Present
695	&& pShwPD->a[i].b.u1Size)
696	{
697	Assert(!(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING));
698	pShwPD->a[i].u = 0;
699	Assert(pPage->cPresent);
700	pPage->cPresent--;
701	}
702	}
703	if (pPage->cPresent == 0)
704	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
705	}
706	goto default_case;
707	#endif /* PGM_WITH_LARGE_PAGES */
708
709	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
710	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
711	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
712	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
713	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
714	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
715	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
716	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
717	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
718	{
719	if (pPage->cPresent)
720	{
721	void *pvShw = PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
722	STAM_PROFILE_START(&pPool->StatZeroPage, z);
723	#if 0
724	/* Useful check for leaking references; very expensive though. */
725	switch (pPage->enmKind)
726	{
727	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
728	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
729	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
730	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
731	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
732	{
733	bool fFoundFirst = false;
734	PPGMSHWPTPAE pPT = (PPGMSHWPTPAE)pvShw;
735	for (unsigned ptIndex = 0; ptIndex < RT_ELEMENTS(pPT->a); ptIndex++)
736	{
737	if (pPT->a[ptIndex].u)
738	{
739	if (!fFoundFirst)
740	{
741	AssertFatalMsg(pPage->iFirstPresent <= ptIndex, ("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
742	if (pPage->iFirstPresent != ptIndex)
743	Log(("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
744	fFoundFirst = true;
745	}
746	if (PGMSHWPTEPAE_IS_P(pPT->a[ptIndex]))
747	{
748	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pPT->a[ptIndex]), NIL_RTGCPHYS);
749	if (pPage->iFirstPresent == ptIndex)
750	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
751	}
752	}
753	}
754	AssertFatalMsg(pPage->cPresent == 0, ("cPresent = %d pPage = %RGv\n", pPage->cPresent, pPage->GCPhys));
755	break;
756	}
757	default:
758	break;
759	}
760	#endif
761	ASMMemZeroPage(pvShw);
762	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
763	pPage->cPresent = 0;
764	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
765	}
766	}
767	/* fall thru */
768
769	#ifdef PGM_WITH_LARGE_PAGES
770	default_case:
771	#endif
772	default:
773	Assert(!pPage->cModifications \|\| ++cModifiedPages);
774	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
775	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
776	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
777	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
778	pPage->cModifications = 0;
779	break;
780
781	}
782	if (!--cLeft)
783	break;
784	}
785	}
786
787	/* swipe the special pages too. */
788	for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
789	{
790	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
791	if (pPage->GCPhys != NIL_RTGCPHYS)
792	{
793	Assert(!pPage->cModifications \|\| ++cModifiedPages);
794	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
795	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
796	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
797	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
798	pPage->cModifications = 0;
799	}
800	}
801
802	#ifndef DEBUG_michael
803	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
804	#endif
805	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
806	pPool->cModifiedPages = 0;
807
808	/*
809	* Clear all the GCPhys links and rebuild the phys ext free list.
810	*/
811	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRangesX);
812	pRam;
813	pRam = pRam->CTX_SUFF(pNext))
814	{
815	iPage = pRam->cb >> PAGE_SHIFT;
816	while (iPage-- > 0)
817	PGM_PAGE_SET_TRACKING(pVM, &pRam->aPages[iPage], 0);
818	}
819
820	pPool->iPhysExtFreeHead = 0;
821	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
822	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
823	for (unsigned i = 0; i < cMaxPhysExts; i++)
824	{
825	paPhysExts[i].iNext = i + 1;
826	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
827	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
828	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
829	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
830	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
831	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
832	}
833	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
834
835
836	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
837	/* Reset all dirty pages to reactivate the page monitoring. */
838	/* Note: we must do this after clearing all page references and shadow page tables as there might be stale references to
839	* recently removed MMIO ranges around that might otherwise end up asserting in pgmPoolTracDerefGCPhysHint
840	*/
841	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
842	{
843	PPGMPOOLPAGE pPage;
844	unsigned idxPage;
845
846	if (pPool->aDirtyPages[i].uIdx == NIL_PGMPOOL_IDX)
847	continue;
848
849	idxPage = pPool->aDirtyPages[i].uIdx;
850	AssertRelease(idxPage != NIL_PGMPOOL_IDX);
851	pPage = &pPool->aPages[idxPage];
852	Assert(pPage->idx == idxPage);
853	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX && pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
854
855	AssertMsg(pPage->fDirty, ("Page %RGp (slot=%d) not marked dirty!", pPage->GCPhys, i));
856
857	Log(("Reactivate dirty page %RGp\n", pPage->GCPhys));
858
859	/* First write protect the page again to catch all write accesses. (before checking for changes -> SMP) */
860	int rc = PGMHandlerPhysicalReset(pVM, pPage->GCPhys & PAGE_BASE_GC_MASK);
861	AssertRCSuccess(rc);
862	pPage->fDirty = false;
863
864	pPool->aDirtyPages[i].uIdx = NIL_PGMPOOL_IDX;
865	}
866
867	/* Clear all dirty pages. */
868	pPool->idxFreeDirtyPage = 0;
869	pPool->cDirtyPages = 0;
870	#endif
871
872	/* Clear the PGM_SYNC_CLEAR_PGM_POOL flag on all VCPUs to prevent redundant flushes. */
873	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
874	pVM->aCpus[idCpu].pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
875
876	/* Flush job finished. */
877	VM_FF_CLEAR(pVM, VM_FF_PGM_POOL_FLUSH_PENDING);
878	pPool->cPresent = 0;
879	pgmUnlock(pVM);
880
881	PGM_INVL_ALL_VCPU_TLBS(pVM);
882
883	if (fpvFlushRemTbl)
884	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
885	CPUMSetChangedFlags(&pVM->aCpus[idCpu], CPUM_CHANGED_GLOBAL_TLB_FLUSH);
886
887	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
888	return VINF_SUCCESS;
889	}
890
891
892	/**
893	* Clears the shadow page pool.
894	*
895	* @param pVM Pointer to the VM.
896	* @param fFlushRemTlb When set, the REM TLB is scheduled for flushing as
897	* well.
898	*/
899	void pgmR3PoolClearAll(PVM pVM, bool fFlushRemTlb)
900	{
901	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, pgmR3PoolClearAllRendezvous, &fFlushRemTlb);
902	AssertRC(rc);
903	}
904
905
906	/**
907	* Protect all pgm pool page table entries to monitor writes
908	*
909	* @param pVM Pointer to the VM.
910	*
911	* @remarks ASSUMES the caller will flush all TLBs!!
912	*/
913	void pgmR3PoolWriteProtectPages(PVM pVM)
914	{
915	PGM_LOCK_ASSERT_OWNER(pVM);
916	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
917	unsigned cLeft = pPool->cUsedPages;
918	unsigned iPage = pPool->cCurPages;
919	while (--iPage >= PGMPOOL_IDX_FIRST)
920	{
921	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
922	if ( pPage->GCPhys != NIL_RTGCPHYS
923	&& pPage->cPresent)
924	{
925	union
926	{
927	void *pv;
928	PX86PT pPT;
929	PPGMSHWPTPAE pPTPae;
930	PEPTPT pPTEpt;
931	} uShw;
932	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
933
934	switch (pPage->enmKind)
935	{
936	/*
937	* We only care about shadow page tables.
938	*/
939	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
940	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
941	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
942	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPT->a); iShw++)
943	{
944	if (uShw.pPT->a[iShw].n.u1Present)
945	uShw.pPT->a[iShw].n.u1Write = 0;
946	}
947	break;
948
949	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
950	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
951	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
952	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
953	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
954	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTPae->a); iShw++)
955	{
956	if (PGMSHWPTEPAE_IS_P(uShw.pPTPae->a[iShw]))
957	PGMSHWPTEPAE_SET_RO(uShw.pPTPae->a[iShw]);
958	}
959	break;
960
961	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
962	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTEpt->a); iShw++)
963	{
964	if (uShw.pPTEpt->a[iShw].n.u1Present)
965	uShw.pPTEpt->a[iShw].n.u1Write = 0;
966	}
967	break;
968
969	default:
970	break;
971	}
972	if (!--cLeft)
973	break;
974	}
975	}
976	}
977
978	#ifdef VBOX_WITH_DEBUGGER
979	/**
980	* The '.pgmpoolcheck' command.
981	*
982	* @returns VBox status.
983	* @param pCmd Pointer to the command descriptor (as registered).
984	* @param pCmdHlp Pointer to command helper functions.
985	* @param pVM Pointer to the current VM (if any).
986	* @param paArgs Pointer to (readonly) array of arguments.
987	* @param cArgs Number of arguments in the array.
988	*/
989	static DECLCALLBACK(int) pgmR3PoolCmdCheck(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PVM pVM, PCDBGCVAR paArgs, unsigned cArgs)
990	{
991	DBGC_CMDHLP_REQ_VM_RET(pCmdHlp, pCmd, pVM);
992	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, -1, cArgs == 0);
993	uint32_t cErrors = 0;
994	NOREF(paArgs);
995
996	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
997	for (unsigned i = 0; i < pPool->cCurPages; i++)
998	{
999	PPGMPOOLPAGE pPage = &pPool->aPages[i];
1000	bool fFirstMsg = true;
1001
1002	/* Todo: cover other paging modes too. */
1003	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1004	{
1005	PPGMSHWPTPAE pShwPT = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
1006	{
1007	PX86PTPAE pGstPT;
1008	PGMPAGEMAPLOCK LockPage;
1009	int rc = PGMPhysGCPhys2CCPtrReadOnly(pVM, pPage->GCPhys, (const void **)&pGstPT, &LockPage); AssertReleaseRC(rc);
1010
1011	/* Check if any PTEs are out of sync. */
1012	for (unsigned j = 0; j < RT_ELEMENTS(pShwPT->a); j++)
1013	{
1014	if (PGMSHWPTEPAE_IS_P(pShwPT->a[j]))
1015	{
1016	RTHCPHYS HCPhys = NIL_RTHCPHYS;
1017	rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[j].u & X86_PTE_PAE_PG_MASK, &HCPhys);
1018	if ( rc != VINF_SUCCESS
1019	\|\| PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[j]) != HCPhys)
1020	{
1021	if (fFirstMsg)
1022	{
1023	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
1024	fFirstMsg = false;
1025	}
1026	DBGCCmdHlpPrintf(pCmdHlp, "Mismatch HCPhys: rc=%Rrc idx=%d guest %RX64 shw=%RX64 vs %RHp\n", rc, j, pGstPT->a[j].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[j]), HCPhys);
1027	cErrors++;
1028	}
1029	else if ( PGMSHWPTEPAE_IS_RW(pShwPT->a[j])
1030	&& !pGstPT->a[j].n.u1Write)
1031	{
1032	if (fFirstMsg)
1033	{
1034	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
1035	fFirstMsg = false;
1036	}
1037	DBGCCmdHlpPrintf(pCmdHlp, "Mismatch r/w gst/shw: idx=%d guest %RX64 shw=%RX64 vs %RHp\n", j, pGstPT->a[j].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[j]), HCPhys);
1038	cErrors++;
1039	}
1040	}
1041	}
1042	PGMPhysReleasePageMappingLock(pVM, &LockPage);
1043	}
1044
1045	/* Make sure this page table can't be written to from any shadow mapping. */
1046	RTHCPHYS HCPhysPT = NIL_RTHCPHYS;
1047	int rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pPage->GCPhys, &HCPhysPT);
1048	AssertMsgRC(rc, ("PGMPhysGCPhys2HCPhys failed with rc=%d for %RGp\n", rc, pPage->GCPhys));
1049	if (rc == VINF_SUCCESS)
1050	{
1051	for (unsigned j = 0; j < pPool->cCurPages; j++)
1052	{
1053	PPGMPOOLPAGE pTempPage = &pPool->aPages[j];
1054
1055	if (pTempPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1056	{
1057	PPGMSHWPTPAE pShwPT2 = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pTempPage);
1058
1059	for (unsigned k = 0; k < RT_ELEMENTS(pShwPT->a); k++)
1060	{
1061	if ( PGMSHWPTEPAE_IS_P_RW(pShwPT2->a[k])
1062	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
1063	&& !pPage->fDirty
1064	# endif
1065	&& PGMSHWPTEPAE_GET_HCPHYS(pShwPT2->a[k]) == HCPhysPT)
1066	{
1067	if (fFirstMsg)
1068	{
1069	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
1070	fFirstMsg = false;
1071	}
1072	DBGCCmdHlpPrintf(pCmdHlp, "Mismatch: r/w: GCPhys=%RGp idx=%d shw %RX64 %RX64\n", pTempPage->GCPhys, k, PGMSHWPTEPAE_GET_LOG(pShwPT->a[k]), PGMSHWPTEPAE_GET_LOG(pShwPT2->a[k]));
1073	cErrors++;
1074	}
1075	}
1076	}
1077	}
1078	}
1079	}
1080	}
1081	if (cErrors > 0)
1082	return DBGCCmdHlpFail(pCmdHlp, pCmd, "Found %#x errors", cErrors);
1083	return VINF_SUCCESS;
1084	}
1085	#endif /* VBOX_WITH_DEBUGGER */

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/VMM/VMMR3/PGMPool.cpp@ 43363

Download in other formats: